Meta AI can reproduce half of Harry Potter book verbatim
SMRTR summary
Meta's LLaMA 3.1 AI model can reproduce significant portions of copyrighted texts, including 42% of the first Harry Potter book. Researchers from Stanford, Cornell, and West Virginia University tested five open-weight models, finding LLaMA 3.1 70B most likely to reproduce text from popular books. The study's methodology involved using 50-token prompts to estimate verbatim reproduction likelihood. These findings complicate the ongoing AI copyright debate, potentially challenging class-action lawsuits against AI companies due to varying results across different texts.
SMRTR provides this summary for quick context. The original article belongs to NewsBytes.
Read the original article